Hi Dave,
case wrote:
> Here is some more info on configuring netcdf on Macintosh:
>
> I now have both 10.5 and 10.6 machines, but for me, the cvs configure step
> fails for both, giving undefined symbols to netcdf routines when compiling
> sander.
>
> After hours of hair-pulling, I think the problem boils down to the $march
> variable introduced in configure, revision 1.37. What happens is this:
> even though $march is empty for macos, the variable passed to the netcdf
> configure is CC="$cc $march", which expands to "gcc ". The extra space
> interferes with netcdf's ability to recognize the type of compiler, and to
> automagically set pgiFortran.
>
> Before this change, netcdf's configure was called with CC="gcc", and
> everything worked.
>
> So, here is my proposed change to configure:
>
> RCS file: /home/amber_cvs/cvsroot/amber11/src/configure,v
> retrieving revision 1.39
> diff -u -r1.39 configure
> --- configure 12 Nov 2009 08:38:47 -0000 1.39
> +++ configure 16 Nov 2009 19:13:31 -0000
> .. -886,9 +886,9 ..
> ./configure $netcdfflag > ../../netcdf_config.log 2>&1
> ;;
> *)
> - env FC="$fc $march" F90="$fc $march" FFLAGS="$ldflags $foptflags" \
> - F90FLAGS="$ldflags $foptflags" CC="$cc $march" LDFLAGS="$ldflags"
> \
> - CFLAGS="$cflags" CXX="$cplusplus $march" CPPFLAGS="-DNDEBUG"
> ./configure $netcdfflag \
> + env FC="$fc" F90="$fc" FFLAGS="$ldflags $foptflags" \
> + F90FLAGS="$ldflags $foptflags" CC="$cc" LDFLAGS="$ldflags" \
> + CFLAGS="$cflags" CPPFLAGS="-DNDEBUG" ./configure $netcdfflag \
> > ../../netcdf_config.log 2>&1
> ;;
> esac
>
>
> I'm reluctant to commit this, however, since something might be depending
> on passing $march to netcdf. Plus, I don't understand why the CVS version
> fails for me but works for other people. So, if Ben and Mengjueh (and
> others) could try out the above change, I would appreciate it.
>
> Note: a "good" netcdf configuration will have the following line in
> amber11/src/netcdf/src/config.log
>
> #define pgiFortran 1
>
> It is *not* really necessary to send -DpgiFortran to configure to get this; at
> least for me, it comes out naturally if the space at the end of the compiler
> is removed. ("Bad" configurations end up with f2cFortran in the #define.)
>
> FWIW: Here are my compilers for Mac 10.5:
>
> gcc (GCC) 4.4.0 20081219 (experimental)
> GNU Fortran (GCC) 4.4.0 20081219 (experimental)
>
> For Mac 10.6:
>
> gcc (GCC) 4.5.0 20090910 (experimental)
> GNU Fortran (GCC) 4.5.0 20090910 (experimental)
For me, the real question is what happens when $march is used in other
parts of the code. Your suggested patch would, I think, create an
inconsistency between netCDF and the rest of Amber.
Does anyone know the best way to configure netCDF with any necessary
"-m" switches? Perhaps by putting those in CFLAGS, FFLAGS, etc.?
Ben
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Mon Nov 16 2009 - 13:00:02 PST
